Anatomical Measures John Ashburner zSegmentation zMorphometry zSegmentation zMorphometry

Contents zSegmentation yGaussian mixture model yIncluding prior probability maps yIntensity non-uniformity correction zMorphometry

Segmentation - Mixture Model zIntensities are modelled by a mixture of K gaussian distributions, parameterised by: xMeans xVariances xMixing proportions zCan be multi-spectral yMultivariate gaussian distributions

Segmentation - Priors zOverlay prior belonging probability maps to assist the segmentation yPrior probability of each voxel being of a particular type is derived from segmented images of 151 subjects xAssumed to be representative yRequires initial registration to standard space

Segmentation - Bias Correction zA smooth intensity modulating function can be modelled by a linear combination of DCT basis functions

Segmentation - Algorithm zResults contain some non-brain tissue zRemoved automatically using morphological operations yErosion yConditional dilation

zBelow: examples of segmented images zRight: some non-brain tissue may be included in the GM and WM classes, which can be removed yAbove: T1 image and “brain mask” yCentre: GM and WM before cleaning up yBelow: cleaned up GM and WM

. Partial volume effects can be problematic - no longer Gaussian Mis-registration with the prior probability images results in poor classification. This figure shows the effect of translating the image relative to the priors before segmenting. Known Problems

Other Limitations zAssumes that the brain consists of only GM and WM, with some CSF around it. yNo model for lesions (stroke, tumours, etc) zPrior probability model is based on relatively young and healthy brains. yLess appropriate for subjects outside this population. zNeeds reasonable quality images to work with yartefact-free ygood separation of intensities

Spatial Normalisation using Tissue Classes zMulti-subject functional imaging requires GM of different brains to be in register. zBetter spatial normalisation by matching GM from segmented images, with a GM template. zThe future: Segmentation, spatial normalisation and bias correction combined into the same model.

Spatial Normalisation using Tissue Classes zThe same strategy as for “Optimised VBM” Original MRITemplate Grey MatterSegment Affine register PriorsDeformation Affine Transform Spatial Normalisation - estimation Spatial Normalisation - writing Spatially Normalised MRI

Contents zSegmentation zMorphometry yVolumes from deformations ySerial scans yVoxel-based morphometry

TemplateWarpedOriginal Deformation Field Deformation field

Jacobians Jacobian Matrix (or just “Jacobian”) Jacobian Determinant (or just “Jacobian”) - relative volumes

Early Late Difference Data from the Dementia Research Group, Queen Square. Serial Scans

Regions of expansion and contraction zRelative volumes encoded in Jacobian determinants. z“Deformations Toolbox” can be used for this. yBegin with rigid- registration

Late Early Warped earlyDifference Early CSFLate CSF Relative volumes CSF “modulated” by relative volumes

Late CSF - Early CSF Late CSF - modulated CSF Smoothed

Voxel-based Morphometry zPre-process images of several subjects to highlight particular differences. yTissue volumes zUse mass-univariate statistics (t- and F-tests) to detect differences among the pre-processed data. zUse Gaussian Random Field Theory to interpret the blobs.

Pre-processing for Voxel-Based Morphometry (VBM)

Units for pre-processed data Before convolutionConvolved with a circleConvolved with a Gaussian Units are mm 3 of original grey matter per mm 3 of spatially normalised space

“Globals” for VBM zShape is multivariate yDependencies among volumes in different regions zSPM is mass univariate y“globals” used as a compromise yCan be either ANCOVA or proportional scaling Where should any difference between the two “brains” on the left and that on the right appear?

Nonlinearity Circles of uniformly increasing area. Smoothed Plot of intensity at circle centres versus area Caution may be needed when looking for linear relationships between grey matter concentrations and some covariate of interest.

Validity of the statistical tests in SPM zResiduals are not normally distributed. yLittle impact on uncorrected statistics for experiments comparing groups. yProbably invalidates experiments that compare one subject with a group. xNeed to use nonparametric tests that make less assumptions. zCorrections for multiple comparisons. yOK for corrections based on peak heights. yNot valid for corrections based on cluster extents. xSPM makes the inappropriate assumption that the smoothness of the residuals is stationary. Bigger blobs expected in smoother regions.

Friston et al (1995): Spatial registration and normalisation of images. Human Brain Mapping 3(3): Ashburner & Friston (1997): Multimodal image coregistration and partitioning - a unified framework. NeuroImage 6(3): Collignon et al (1995): Automated multi- modality image registration based on information theory. IPMI’95 pp Ashburner et al (1997): Incorporating prior knowledge into image registration. NeuroImage 6(4): Ashburner et al (1999): Nonlinear spatial normalisation using basis functions. Human Brain Mapping 7(4): Ashburner & Friston (2000): Voxel-based morphometry - the methods. NeuroImage 11: References